Nvidia's Balance Sheet Is Now Part of the Pitch as AI Infrastructure Bets Multiply

Nvidia underwrites $105B for OpenAI's Ohio data center, invests in power infrastructure, and joins $500B AI funding coalition—shifting risk to construction and execution.

The line between chipmaker and financier has all but disappeared at Nvidia. Over the past week alone, the company has committed to underwriting a colossal OpenAI data center complex in Ohio, taken a minority stake in a Houston-based power infrastructure developer, and helped assemble a coalition of investors targeting more than half a trillion dollars in AI buildout funding. For shareholders, the question is no longer just how many chips Nvidia can sell — but whether the projects it is helping to fund will actually deliver.

The most striking commitment came on Monday, when Nvidia disclosed it would provide up to $105 billion in credit for the OpenAI facility in Ohio. The line initially covers 4.25 gigawatts of computing capacity, with an option to expand by another 3.75 gigawatts. Nvidia is not merely supplying capital here; the compute itself will be its own hardware, with capacity scheduled to come online in stages starting in 2028. The company is also investing $1.5 billion directly in SB Energy, the developer behind the project, to accelerate its transformation into a major AI infrastructure player.

Elon Musk was quick to cast doubt, posting on X on Tuesday that the project would take "much longer to come online than the people involved think." That skepticism touches on a genuine shift in risk profile. When Nvidia moves from selling chips to financing the sites where those chips run, it takes on construction timelines and execution risks that sit well outside its traditional control.

A widening web of commitments

The Ohio deal is hardly an isolated move. On August 17, Nvidia filed a regulatory notice with the SEC covering a guarantee for SB Energy's PORTS-Pike Technology Campus — an eight-gigawatt data center site in Ohio dedicated exclusively to Nvidia-powered compute for OpenAI. Days earlier, on August 10, the company joined forces with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish independent financing platforms aimed at mobilizing more than $500 billion for AI infrastructure.

There are also smaller but telling positions accumulating across the portfolio. Nvidia recently announced a long-term partnership with Ilya Sutskever's startup Safe Superintelligence, tied to a $5 billion investment. Its stake in SpaceX, acquired through an investment in xAI, was worth roughly $21 billion at the end of the second quarter — the second-largest holding behind its Intel position, which had slipped to about $22 billion from $30 billion at the start of the quarter.

Regulators are smoothing the path. The SEC has issued an administrative directive easing securitization rules for data center debt, effectively exempting Nvidia's $500 billion financing initiative from the risk-retention requirements of Dodd-Frank. That means the massive capital-raising effort will proceed in a friendlier regulatory environment than initially anticipated.

A quieter stock, a louder earnings test

The market's reaction so far has been measured. The shares were trading around €184.26 on Friday, down 0.7 percent on the day and 5.4 percent lower on the week, though still up 15 percent since the start of the year. The stock sits about 9 percent below its 52-week high of €202.50, reached in May.

Some of that caution may reflect the sheer scale of what Nvidia is now promising. The company has effectively become creditor, investor and co-owner of an ecosystem it once simply supplied. Short term, that secures future chip demand. Longer term, the returns depend on whether the financed projects actually generate the promised yields.

The real test arrives on August 26, when Nvidia reports results for its second fiscal quarter of 2027, which ended in July. Management has guided to revenue of approximately $91 billion, plus or minus 2 percent. Jefferies analysts expect the company to blow past that figure, forecasting a record $95 billion against a market consensus of $92.07 billion. The same team projects that the upcoming Vera Rubin system could account for more than 40 percent of GPU revenue by the fourth fiscal quarter of 2027.

BMO Capital initiated coverage on Friday with an "Outperform" rating and a $340 price target, calling Nvidia its preferred pick among large-cap semiconductor names.

None of this changes the underlying tension. A new dividend of 25 cents per share and the $80 billion buyback program approved in May are welcome gestures, but they do little to address the central question: whether the data centers Nvidia is now helping to finance come online on time and turn a profit. Until then, the stock is likely to keep swinging between enthusiasm for the opportunity and wariness about the balance-sheet exposure that comes with it.

Meanwhile, Nvidia has pushed back against a report from The Information claiming the company would launch a China-specific language processor (LPU) by year-end, stating that no such product is on the roadmap. The denial coincides with reports that ByteDance and Tencent each received around 10,000 H200 processors after Chinese regulators approved small import volumes despite US export controls.